CXMT IPO: A Ceiling Behind the Windfall Profits
Claire Weston
CXMT is pursuing China's largest semiconductor IPO in years, targeting roughly RMB 29.5 billion in proceeds after revenue surged 156% in 2025 and net profit topped $1 billion for the first time — but the profit explosion is driven by a once-in-forty-years DRAM shortage cycle, not a market-share breakthrough.
How did CXMT get here?
CXMT was founded in 2016. Its technical DNA traces to Qimonda, a bankrupt European memory giant — CXMT acquired patents, technical documentation, and key engineers to bootstrap from zero.
The Hefei municipal government absorbed over RMB 36.6 billion in cumulative losses across nearly a decade, acting as patient capital. This means → without sustained state backing, this company would not have survived to reach profitability.
CXMT is now the world's fourth-largest DRAM (dynamic random-access memory — the most common memory chip in phones and servers) manufacturer.
How real are these profit numbers?
2025 revenue hit roughly $8.6 billion, up 156% year-on-year; net profit exceeded $1 billion for the first time. Q1 2026 revenue reached $7.3 billion with an operating margin of about 70%.
In plain terms = the real driver is a "once-in-forty-years shortage" in global DRAM that sent average prices soaring. CXMT's own market share did not materially expand, and its cost structure did not fundamentally improve.
This reflects a classic cyclical-stock pattern: margins look spectacular when prices spike, but they can contract just as sharply when the cycle turns.
Where will the IPO proceeds go?
The IPO targets roughly RMB 29.5 billion (about $4.1 billion): nearly 70% earmarked for wafer-fab lines and DRAM technology upgrades, the remaining 30% for forward-looking R&D.
The prospectus makes no mention of a dedicated HBM investment plan. This means → on the hottest AI-memory track in the industry, CXMT has no explicit capital commitment at the fundraising level.
On governance: CXMT uses concert-party agreements to concentrate voting control in a minority economic stake. Consolidated accounts overstate public shareholders' effective claim to a degree. Post-listing, state and policy-linked capital will still hold over 30%; Alibaba Cloud is both a major customer and a nearly 4% shareholder, providing structural demand support for domestic shipments.
Where is the capacity ceiling?
CXMT projects monthly capacity of 350,000 wafers by end-2026, rising to 500,000 by end-2028 — roughly 17% of global DRAM supply at that point.
But the ceiling is already visible. U.S. restrictions on EUV lithography tools — equipment that etches chip circuits with extreme-ultraviolet light — have been in place since 2019, reinforced by additional export controls in 2022 and 2024. This directly blocks CXMT's path to advanced nodes below 18 nm.
In plain terms = capacity can keep scaling, but the process node cannot keep shrinking. Volume grows, yet per-unit competitiveness may weaken over time.
HBM — how does CXMT close its biggest technology gap?
Nearly 100% of CXMT's sales come from conventional LPDDR and DDR products. In HBM — high-bandwidth memory, the essential companion chip for AI training accelerators — the company is virtually absent.
HBM3 front-end yield sits at roughly 35%, back-end yield at about 70%, giving a combined yield of only ~25%. This means → three out of every four chips fail; mass production is still a considerable distance away.
CXMT's response: skip HBM3 entirely, pivot to 8-layer and 12-layer HBM3E development, and partner with Huawei on a custom architecture independent of the JEDEC standard.
How far along is supply-chain localization?
Etch, deposition, and polishing steps already use domestic equipment from AMEC, Piotech, NAURA, and HWATSING — localization is accelerating in these segments.
But for lithography, ion implantation, and TSV (through-silicon via — the process that connects chip layers vertically), CXMT remains heavily dependent on pre-restriction tools from ASML, Applied Materials, and Lam Research. Any disruption to maintenance or upgrades would hit capacity directly.
This reflects CXMT's central contradiction: short-term profits are tightly bound to a cyclical pricing windfall, while long-term competitiveness depends on cracking the HBM yield bottleneck and achieving supply-chain self-sufficiency under export controls — and neither of those gates has a certain answer today.
Content is for reference only, not financial advice.